Differences in glycine receptor turnover and synaptic size distributions between two functionally distinct classes of motoneurons in larval zebrafish

Summary

The interplay between binding and unbinding of synaptic receptor proteins at synapses plays an important role in determining receptor concentration and synaptic strength, with known links between changes in binding kinetics and synaptic plasticity. The regulation of such kinetics may subserve the specific functional requirements of neurons in intact circuits. However, the majority of studies of synaptic turnover kinetics have been performed in cultured neurons outside the context of normal circuits, and synaptic receptor turnover has not been measured at individual synaptic sites in vivo. We quantified the distribution of glycinergic receptor dynamics using fluorescence recovery after photo-conversion of synapses in intact zebrafish, and correlated recovery kinetics to synaptic volume in two functionally distinct classes of cells: primary and secondary motoneurons. The rate of fluorescence recovery after photo-conversion decreased with synaptic volume in both types of motoneurons, with larger synapses having slower recovery. Primary motoneurons had both larger synapses and associated slower recovery times than secondary motoneurons. Our results suggest that synaptic kinetics are regulated in concert with synaptic sizes and reflect the functional role played by neurons within their circuit.

Introduction

Despite the long-term stability of synaptic structures, their molecular constituents are dynamic. Postsynaptic receptor and scaffolding proteins diffuse freely in the peri-synaptic membrane and can be transiently trapped at synaptic sites [1,2]. Plasticity occurs in part via changes in the number of trapped synaptic proteins [3,5]. Therefore, the dynamic equilibrium between binding and unbinding of synaptic components influences receptor numbers and synaptic strength [6–10]. These synaptic kinetics are regulated over short time scales by activity dependent processes through post-translational modifications [1,11–14]. On longer time scales, homeostatic synaptic scaling is also associated with changes in diffusional dynamics of receptors [9,10,15–17].

The well-established importance of receptor turnover kinetics in synaptic dynamics over short and long time scales points to critical roles for such turnover in vivo. We might expect differences in patterns of turnover at different synapses on individual neurons, as well as differences between cell types that might reflect their functional roles in networks. Most studies of receptor turnover have, however, focused on cultured neurons. Receptor turnover at individual synapses in intact animals has not been measured, although the ability to do it would place synaptic properties in the context of intact circuits. Here we measure the kinetics of turnover of glycinergic receptors at individual synapses on different classes of spinal motoneurons in intact larval zebrafish to explore how turnover varies across synapses and functionally different neurons in vivo.

Zebrafish motoneurons form a functionally heterogeneous population that is recruited in a systematic way as swimming speed increases. Smaller, so-called secondary motoneurons (SMNs) in each body segment are involved in slower speed swimming, while individually identifiable primary motoneurons (PMNs) are recruited only for the fastest swimming movements and escapes [18–21]. The neurons develop sequentially beginning with the PMNs, which drive strong, coarse movements, followed later by SMNs which control weaker and lower amplitude behaviors [21,22]. This orderly addition is associated with a developmental refinement of movement control, offering the possibility of exploring how receptor turnover relates to functional subdivisions within a cell type.

The patterned activation of all types of motoneurons depends on glycinergic synapses, which are essential for control of the alternating bends during swimming, and are known to show homeostatic regulation in vivo in zebrafish [23]. We used multiphoton imaging and fluorescence recovery after photo-conversion of dendra2-tagged glycine receptor (GlyR) α1 subunit to determine the kinetics of GlyR turnover at synapses in intact zebrafish. We then tested the hypothesis that the functional differences between PMNs and SMNs are associated with differences in the kinetics of receptor turnover in the population of synapses on the two classes of cells. Our work reveals differences in the kinetics of glycinergic synapses on PMNs and SMNs, with PMNs having larger synapses with slower turnover of receptors. The difference may relate to the functional roles of the cells, as more dynamic regulation of synaptic strength in SMNs might allow for constant, subtle refinements of motor behavior resulting in the more refined graded control possible from the summation of weaker motor units. Our work raises the possibility that differences in receptor turnover among neurons may support their roles in circuits by tuning the lability of synapses and thus plasticity to match neurons to their function.

Results

Dendra2-Tagged GlyRα1 localizes at glycinergic synapses and can form functional channels

In order to examine the kinetics of GlyR turnover we stochastically expressed GlyRα1 subunit tagged with the photo-convertible fluorophore dendra2 in PMNs, via injection into single-cell stage zebrafish embryos of Gal4 driven by the vesicular acetylcholine transporter (VAT) promoter along with a plasmid containing UAS-Dendra2-GlyRα1. This produced labeled cholinergic neurons [24] sparsely distributed in the spinal cord and brain. Dendra2-tagged receptors formed discrete puncta (Fig. 1A) on the somatic regions, ventral dendrite, and dendritic arbor of PMNs. A large cluster of receptors was evident at a known glycinergic synaptic site on the ventral dendrite of the motoneuron, where it receives potent commissural inhibition [25–27]. These observations and the overall distribution of clusters were consistent with synaptic targeting of receptors. To test the synaptic localization of the GlyRα1 more carefully, we used double immunostaining for a similarly constructed GlyRα1-GFP (due to the availability of anti-GFP) and the inhibitory synaptic marker gephyrin, to ask whether the GlyRα1-GFP was located at sites also containing endogenous gephyrin. The two were well co-localized, with an average 73% of automatically determined [4] above-threshold GFP voxels also having above-threshold gephyrin staining. The two channels had a mean correlation of r=0.35 in synaptic regions, suggesting that GFP and gephyrin fluorescence co-varied. These results are consistent with the transgenic GlyRα1 construct being trafficked to synaptic sites in vivo (Fig. S1).

Figure 1. The expression of dendra2-GlyRα1 in motoneurons, physiology of dendra2-GlyRα1 channels in muscle fibers, and synapse level targeting for dendra2 photo-conversion.

(A) A single motoneuron expressing dendra2-GlyRα1 (green) along with membrane targeted tdTomato (magenta), exhibiting distributed glycinergic puncta along the labeled soma and dendrites. (See also Figure S1 for co-staining with gephyrin). (B) Expression of dendra2-GlyRα1 in muscle to test channel formation. Left: Transgenic muscle fiber targeted for patch recording. Right: Voltage clamp recordings at the indicated holding potentials in a muscle fiber. (C) Precise targeting with a 405 nm laser allowed us to convert synaptically localized dendra2-GlyRα1 from green (top) to red fluorescence (bottom, represented as magenta) with near single-synapse accuracy.

We tested whether the tagged construct could produce a conducting channel by expressing it in muscle fibers, which do not natively express GlyRs. Because GlyRα1 forms functional homomeric channels in vitro [28] we expected that functional dendra2-tagged channels would be able to conduct chloride. We expressed dendra2-GlyRα1 using the CMV promoter and performed whole-cell patch recording from muscle fibers (Fig. 1B). Dendra2 positive fibers exhibited currents with a reversal potential between −45 and −55 mV (N=3) in response to pulses of 10 mM glycine, consistent with a chloride current given our physiological solutions (Fig 1B). The labeled fibers did not respond to pulses of extracellular solution without added glycine, and unlabeled muscle fibers were non-responsive to 10 mM glycine (not shown). We conclude that the construct was able to form conducting channels.

Fluorescence recovery after photo-conversion of individual synaptic sites reveals a correlation between GlyR turnover kinetics and synaptic size

Synaptically bound GlyRs and GlyR-gephyrin complexes may become disassociated from synaptic sites and diffuse laterally in the plasma membrane [10,29–31]. We used fluorescence recovery after photo-conversion at synapses to estimate the unbinding kinetics of GlyRs. By focally converting synaptically bound dendra2-GlyR from green to red and then tracking converted synapses over time we were able to quantify the dynamics of GlyR turnover in vivo. We used precise multiple low-power bursts from a 405 nm laser to photo-convert one or more synapses on labeled PMNs (Fig. 1C). Converted synapses exhibited a marked increase in red fluorescence (12.7±1.5 fold vs. non-converted synapses, N = 234) and a marked decrease in green (to 21.2± 0.75% of pre-conversion intensity, Fig. S5E). We initially monitored neurons at 30–50 minute intervals for up to 3 hours, and at intervals of several hours thereafter (Fig 2 and S2). Over 28 hours, repeated imaging and tracking of a single converted synapse revealed that green fluorescence recovered over time (Fig. S2A). However, the overall intensity of labeling sometimes fluctuated as might be expected due to factors such as bleaching (Fig. S2A inset and Fig. S2B), the photo-conversion process, movement of the larvae, and variations in embedding depth and angle for longer time series. We corrected for these broad intensity changes by normalizing the green fluorescence intensity of converted synapses at each time point by the mean of non-converted synapses. In imaging experiments lasting more than 10 hours, the average level of recovery (after correction) was 103±11.3% of the original fluorescence (N=15 synapses), and in experiments between 6 and 10 hours it was 57.1±5% (N=24 synapses). Corrected time-courses for different synaptic sites exhibited heterogeneous recovery times (Fig. 2B and C) that were fit well with exponential recovery curves of the form f(t) = 1-exp(−t/ τ) (Fig. 2C).

Figure 2.

Analysis of photo-conversion recovery time courses. (A) The normalized maximum projection time series photo-converted synapses in 5 different PMNs and estimated recovery constant τ. (B) The data from A plotted along with estimated recovery curves. (See also Figure S2).

Next we measured the distribution of recovery constants sampled from 17 PMNs at 8 days post-fertilization (8dpf), an age at which spinal locomotor circuits are functional. The recovery times varied (median τ=8.8 and interquartile range of 5.1–17.0 hours) and appeared to be associated with synaptic size. In contrast, there was no significant correlation between the effectiveness of conversion as quantified by the ratio of post and pre-conversion intensities, and τ (not shown, P=0.25). A plot of τ against the relative starting intensity of the synapses revealed that larger synapses had longer recovery times (Fig. 3A). Because we converted multiple synapses per cell, we constructed a linear mixed effects (LME) model for ln τ (to satisfy the assumption of normality) which controls for repeated cell contributions. Our model supports the hypothesis that a larger starting size significantly increases the average value of ln τ. Examining whether the converted synapses were located on the dendritic arbor or on the ventral dendrite/axon hillock, from which the motor axon arises, revealed that both synapse types followed the same size/recovery trend (Fig. 3B) with ventral dendrites tending to have larger size and slower recovery kinetics (ln τ = β1*Relative Intensity + β2, β1 = 0.3866 ln h, std. err. = 0.10, P<0.001; β2=0.3941 ln h for axonal synapses and 0 otherwise, std. err. = 0.23, P=0.09).

Figure 3. GlyR turnover kinetics in 8dpf primary motoneurons.

(A) The natural log of the estimated recovery parameter τ of 77 synapses sampled from seventeen 8dpf primary motoneurons (each color represents photo-converted synapses from the same cell, with X marking the median for that cell) plotted against its normalized beginning intensity (relative intensity). (B) The same data from the top plot re-coded to illustrate synapses present on the dendritic arbors (red) or on the ventral dendrite/axon hillock region (black).

In summary, we sampled the rates of GlyR turnover of PMN synapses in intact larval zebrafish. We found that PMN synapses exhibit a range of τ spanning 2 hours to more than 17 hours. Furthermore, our analysis reveals that more intense, larger synapses have slower turnover kinetics. Consistent with this general trend, synapses on the ventral dendrite/axon hillock regions where especially powerful glycinergic synapses are located [25–27] have slower off kinetics than other dendritic locations.

Secondary motoneurons have faster turnover kinetics than primary motoneurons

The PMNs represent one class of motoneurons involved in powerful movements. Another class, secondary motoneurons (SMNs), participates in slower motor behaviors [19,32]. We sought to explore possible correlations between kinetics and the functional role played by each motoneuron type, as kinetics might reflect differences in synaptic size or plasticity. To control for animal age differences and also explore how kinetics change over the course of development, we sampled a variety of larval ages. In some cases, we tracked the same cells over time. PMNs and SMNs are easily distinguishable based on axonal morphology, soma size, and position in spinal cord (Fig. 4A) [21,33]. Measuring the kinetics of 94 PMN synapses in 20 PMNs and 77 SMN synapses in 15 SMNs, we found that, although their distributions overlapped, the distribution of recovery times in PMNs was significantly shifted toward longer times (Fig. 4B, two sample KS-test, D=0.3349, P<0.0001). Furthermore, we observed fast turnover rates in SMNs that we did not encounter in our initial data set from 8dpf PMNs. Similar results were obtained through an analysis of the decay of normalized red fluorescence (Fig. S3A) and were significantly correlated with the kinetics of green fluorescence recovery (Fig. S3B, β=0.7440, t(170)=31.3, P<0.001). Though well correlated, the values from analysis of the red were systematically shifted lower than the green ones at individual synapses, possibly as a result of differences in normalization procedure and penetration depth for the two wavelengths. We wondered whether faster recovery kinetics in SMNs could be explained by differences in synaptic sizes or locations, and so next sought to examine differences in synaptic organization between the two cell types.

Figure 4. The distribution of GlyR turnover kinetics measured from primary and secondary motoneurons.

(A) Images of a 9dpf denra2-GlyR α1 labeled PMN (left) and SMN (right) prior to photo-conversion experiments. PMNs have larger somas than SMNs and have large specialized glycinergic synapses on the ventral dendrite/axon hillock region (see ventral cluster arrow). (B) The overlaid histograms of estimated τ (on log axis) from photo-conversion experiments on 94 PMN synapses (gray) and 77 SMN synapses (green). PMN synapses were more likely to have long recovery times (larger values of ln τ) than SMN synapses (2 sample KS-test, P<0.001). (See also Figure S3).

Primary motoneurons have larger synapses than secondary motoneurons

Because slower kinetics within PMNs are correlated with larger synaptic size (see Fig. 3), one possible explanation for the difference in kinetics we observed across cell types is that mean synaptic size is greater in PMNs than in SMNs. However, our previous metric of synaptic size, relative intensity, could not be used to compare between different preparations as it was normalized within each preparation. Furthermore, raw measurements of intensity cannot be used because of variation with excitation intensity and preparation opacity, in addition to receptor numbers. In order to circumvent these issues, we developed and validated an alternative synaptic volume estimation method (Fig. S4). Briefly, multi-photon stacks were de-convolved and then thresholded at 6σ above the mean intensity of dendritic regions. Synaptic volume was then estimated as the volume of above-threshold voxels in a defined cuboid enclosing each particular synapse (see Supplemental Experimental Procedures for a more detailed explanation).

We then applied this technique across cell types to determine whether PMNs and SMNs show differences in their synaptic volumes. In 8–9dpf animals (N=14 PMNs, 11 SMNs), we observed that SMNs have smaller mean synaptic volumes than PMNs (Fig. 5A left, x̄ = 0.23 μm³ for SMNs and 0.45 μm³ for PMNs, P<0.05, Student’s T-Test). Consistent with this, the mean volume of the largest 10% of synapses per cell was also significantly greater in PMNs than in SMNs (Fig 5A right, x̄ = 0.61 μm³ for SMNs and 1.45 μm³ for PMNs, P<0.01, Student’s T-Test). However, mean synaptic volume also decreased as a function of distance from the cell soma in both cell types, as did the average frequency of synapses (Fig. 5B). We constructed a LME model for volume. (transformed to satisfy normality, see Fig 5C) in order to control for these factors. In the model, where each β describes the rate of change of transformed volume, there is a significant effect of both cell type (β = −.10 μm³ decrease for SMNs, std. err. = 0.05, P<0.05) and synaptic distance to soma (β = −.003 μm³ per μm distance, std. err. = 0.0004, P<0.001). These results indicate that at 8–9dpf, the distribution of synaptic volumes in PMNs has a significantly larger mean than in SMNs, and that synaptic volume decreases as a function of distance from the cell soma in both cell types. Given our earlier evidence that turnover rate varies with synapse size, the differences between PMNs and SMNs might be a result of differences in overall synaptic sizes.

Figure 5. Synaptic volumes and synaptic distances to soma measured in 8–9dpf motoneurons.

(A) Box-and-whisker plots of the median (red line), interquartile (box) and full range (whiskers) of data along with automatically determined outliers (+ symbols) for the distribution of mean synaptic volumes of all synapses per neuron (left) or the largest 10% (right) of 14 PMNs (1239 synapses) and 11 SMNs (665 synapses) in 8–9dpf animals. A two-sample t-test confirmed a statistically significant difference in the overall distributions (medians 0.23 μm³ and 0.45 μm³, * P = 0.017) and the 10% tail (medians 0.60 μm³ and 1.45 μm³, ** P = 0.004) between PMNs and SMNs. (B) The average synaptic volume (bars, left abscissa) and average number of synapses (lines, right abscissa) at increasing dendritic distances to soma plotted for the same set of PMNs and SMNs. (C) The average frequency histogram of synaptic volume for the same PMNs and SMNs after 4^th root transformation to produce normality. The non-transformed average cumulative distribution for each cell type is shown in the inset. (See also Figure S4).

Synaptic size and synaptic distance to soma explain variation in GlyR turnover kinetics

To explore the potential contribution of synaptic size versus other variables to turnover kinetics, we first performed exploratory statistics on our photo-conversion sample to see how parameters varied between the two cell types (see Supplementary Experimental Procedures and Fig. S5) and chose significantly predictive variables to include in an LME model for 3–14dpf motoneurons (ln τ ~ Synaptic Volume + Synaptic Distance to Soma + Larval Age + Cell Type + Larval Age * Cell Type). Although kinetics varied overall with dorsal-ventral soma position of MNs, this factor was not included in the model because the correlation within each cell type was not significant (Fig. S5D). As before, there was no significant correlation of ln τ to the effectiveness of photo-conversion (Fig. S5E.) Ln τ increased significantly with synaptic volume, indicating slower turnover kinetics at larger synapses. (Fig. 6, left; β=0.38 ln h per μm³, std. err. = 0.09, P<0.001). Similarly, kinetics slowed as synaptic distance to soma increased (Fig 6 right, β=0.03 ln h per μm distance, std. err. = 0.008, P<0.001). There was a small interaction between larval age and motoneuron type such that SMN synapses exhibited slower turnover rates with increasing larval age while PMN synapses did not (SMN ln τ = 0.09 ln h * larval age dpf – 1.04 by comparison to a PMN synapse of the same volume and distance to soma, std. err. = 0.04, P<0.05). Overall, this model emphasizes the importance of synaptic volume and distance from the soma in the prediction of GlyR turnover kinetics. Other factors being equal, larger synapses exhibit slower turnover than smaller synapses, and synapses distal to the soma have slower kinetics than those that are proximal for a given size synapse. Furthermore, this result suggests that differences in the distribution of observed kinetics between PMNs and SMNs are primarily due to differences in synaptic volume.

Figure 6. Relationships between τ and synaptic volume or synaptic distance to cell soma.

(A) τ values for photo-converted PMN and SMN synapses plotted against their volumes. The particularly large synapses enclosed in the dashed box are specialized synapses on the ventral dendrite/axon hillock region of PMNs. (B) The τ of all photo-converted synapses for PMNs and SMNs plotted against their distances along the dendrite to the cell soma. Note that τ is plotted on a log axis for both panels. (See also Figure S5 and S7).

Young Synapses are Smaller and Have Faster Turnover Kinetics than Older Synapses

Our data indicates that differences in GlyR turnover kinetics between PMNs and SMNs are a result of differences in the distribution of synaptic volumes between the two cell types. It is unclear, however, how this difference arises. One hypothesis is that PMNs have a longer time to develop than SMNs and thus have larger synapses. However, an analysis of age-matched MNs did not support this view (see Fig. S6B). Another possibility is that PMNs and SMNs have synapses of differing ages. In order to examine the relationship of our previous results to synaptic age as opposed to motoneuron or larval age, we imaged neurons at 5dpf, 7dpf, 8dpf, and 9dpf and identified the appearance and disappearance of inhibitory synapses between consecutive time points (Fig. 7A). We first sought to explore the relationship between synaptic age and volume by comparing the sizes of synapses at 8dpf (Fig. 7Bi) and 9dpf (Fig. 7Bii) that arose at known days before then. At both ages, older synapses were larger than younger ones in PMNs (Volume. = Synapse Age + Type + Synapse Age*Type; βAge_8dpf = 0.09 μm^3/4 per day, std. err. = 0.01, P_8dpf<0.001; βAge_9dpf = 0.065 μm^3/4 per day, std. err. = 0.008, P_9dpf < 0.001), where synapse ages were modeled as categorical variables. However, synaptic volume for SMNs increased significantly with age only in the 9dpf dataset (βAge_8dpf = −0.004 μm^3/4 per day; std. err. = 0.02, P_8dpf<0.001; βAge_9dpf=0.065 μm^3/4 per day, std. err. = 0.008, P_9dpf<0.001). This difference may be the result of SMN synapses reaching a smaller ultimate size than PMN synapses and thus not increasing linearly in volume after about 3 days (note that 9dpf motoneurons have an additional category of 2–3 days in Fig. 7B).

Figure 7. Synaptic age and kinetics.

(A) Example PMN (left) and SMN (right) imaged at 5, 7, 8, and 9dpf were examined for the formation of new synapses. Individual synapses are numbered to show examples of tracking over time. Note the appearance of new synapses 9 and 10 at 8dpf on the PMN (left, blue inset) due to the growth of a new branch from a more ventral process and the loss of the branch containing synapses 5, 6, and 7 previously. White bars in bottom left of panels are 2 μm. (B) Synaptic volumes on 8dpf (top) and 9dpf (bottom) motoneurons organized by the timing of formation for each synapse (N=8 PMNs and 7 SMNs total, see text for statistics; see also Figure S6). (Ci) Cumulative frequency histograms for τ of new, 1 day, and 2+day old synapses. (Cii) The τ (on log axis) of photo-converted synapses plotted against their volume. The blue line represents the fit of a LME model to 2+ dpf synapses. (See also Figure S7).

To test the possibility that young synapses might be less stable and have faster turnover kinetics than older synapses, we performed a photo-conversion experiment at 8dpf on synapses that were newly formed (appeared at 8dpf), 1 day old (appeared at 7dpf), or 2+ days old (present at 5dpf). New synapses had significantly faster kinetics than older ones (Fig. 7Ci). To explore the effects of synaptic age on kinetics more systematically, we constructed a LME model (ln τ = Synaptic Volume + Synaptic Distance to Soma + Synaptic Age + Cell Type + Synaptic Volume * Synaptic Distance to Soma) with synaptic age as a categorical variable. Consistent with previous results, volume significantly contributed to ln τ with bigger synapses having longer τ (β=2.41 ln h per μm³, std. err. = 0.82, P<0.005). Similarly, turnover rate decreased with increased synaptic distance from the soma (β=0.029 ln h per μm, std. err. = 0.014, P=0.052). The interaction term between distance from soma and volume was not significant (P=0.08). Our data indicates that ln τ is similar for a synapse of the same volume and synaptic distance to soma regardless whether located in a PMN or SMN (P=0.17). In both PMNs and SMNs, synapses that appeared at 8dpf tended to have faster kinetics than older synapses (Fig. 7Cii, β= −.66 ln h in new synapses compared to 2+ day old synapses, std. err. = 0.24, P=0.008). Our model suggests that independent of synaptic volume and synaptic distance to soma, newly formed synapses have lower ln τ values than older ones. Together, these results lead to the conclusion that newly formed synapses may be more labile than more established ones and exhibit increased rates of GlyR unbinding.

In summary, we have quantified the kinetics of GlyR turnover in vivo for the first time in PMNs and SMNs and demonstrated that they exhibit a distribution of τ values ranging from less than an hour to tens of hours. PMNs exhibit a tail of slower recovery times in their synaptic kinetics that is not present in SMNs. Our LME models show that synaptic volume and synaptic distance to soma are key determinants of GlyR kinetic rates in both cell types. Furthermore, comparisons of overall differences in the distribution of synaptic volume between the two cell types at 8dpf leads to the conclusion that PMNs have large synapses that are not present in SMNs, and indicates the differences in kinetics between PMNs and SMNs are due to differences in the distribution of synaptic volumes. Although synapses increase in volume with age, SMN synapses may reach a smaller average size than PMN synapses. Lastly, newly formed synapses have faster kinetics than more established ones, suggesting that kinetics reflects in part the stability of synapses.

Discussion

The distribution of synaptic sizes and postsynaptic receptor dynamics are critical for neuronal function. The two are linked, as the balance of receptor binding and unbinding at synapses determines the steady-state number of receptors and synaptic strength. We might predict differences in synaptic size distributions and associated receptor dynamics between neuronal classes based on their functional roles, but such a relationship had not previously been studied. Our ability to monitor turnover at particular synapses in vivo allowed us to explore the relationship between turnover kinetics at glycinergic synapses of different sizes and ages located at different places on a dendritic arbor. We did this for two functionally different cell types, primary and secondary motoneurons, to explore how receptor dynamics and synaptic size distributions differ in neurons driving fast versus slower motor outputs.

The rates in vivo for primary motoneurons (mean τ of 11.8 hours) were on the order of membrane-bound GlyR turnover measured in vitro (mean τ of 20.2 hours), from studies of picrotoxin-blocked ventral horn neurons in culture [29]. There are no in vivo studies of receptor turnover at individual central synapses. Turnover of acetylcholine receptors (AChRs), however, has also been investigated in vivo at the neuromuscular junction and within the submandibular ganglion [34,35]. The turnover rates observed for AChRs in the SMG matches the faster PSD95 turnover rates measured at cortical synapses (τ = 57.7 minutes) [34], but stands in contrast to much longer turnover rates observed at the NMJ (τ = 484.7 hours). In both cases, the toxin used to label receptors may have altered receptor turnover rates from basal levels.

GlyR off kinetics decline with increasing synaptic size in both classes of motoneurons we studied. A similar relationship exists for the synaptic scaffolding protein PSD-95 in excitatory cortical spines [8] . PSD-95 turns over at much faster rates than GlyR, but its turnover rates were also slower in larger spines. Such a relation between size and turnover rates may in part result from a lower ratio of circumference to area in large synapses. As rates of recovery after photo-conversion reflect unbinding (off) rates in cases where binding kinetics are much slower than diffusion, as is the case here [10,17,36,37], the data suggest that differences in receptor levels may be regulated in part by unbinding rates. Importantly, the work in PSD95 and AChRs revealed changes in kinetics after deprivation of synaptic input [8,35] or axotomy [34] consistent with activity-dependent processes altering strength via changes in binding affinity.

In both PMNs and SMNs in 8–9dpf larval zebrafish, the number of glycinergic synapses decreased with distance from the soma and this was accompanied by a small, but significant decrease in synaptic size. The decline in glycinergic synapse numbers is similar to what is seen in mammalian motoneurons. However, the decrease of glycinergic synaptic size with distance along the arbor differs from the situation in adult mammalian spinal cord, where the average cluster size and morphological complexity of glycinergic synapses onto α MNs and type Ia interneurons increases as a function of distance, possibly to increase the influence of distal synapses. Renshaw cells exhibit a dramatically different organization, having much larger synapses on the soma and proximal dendrites than MNs or type Ia interneurons, and a complete absence of glycinergic synapses at distal locations 500 μm or greater from the soma [38]. However, our evidence supports the conclusion that the turnover rate at proximal and distal synapses of similar sizes is slower at more distal synapses. We speculate that this may be a consequence of decreased abundance of free membrane-bound receptors distally.

Our ability to image the same neuron repeatedly over days allowed us to observe when synapses arose and track their size and receptor turnover kinetics over time. Synapses increased in size with synaptic age in both PMNs and SMNs (Fig. 7B). Although our sample of new synapses was limited, our analysis indicates that the newest, smallest synapses have faster turnover kinetics than predicted based on the overall relationship between size and kinetics in our data sets (7C). Synapses older than one day, however, had the same size/kinetics relationship in both PMNs and SMNs, suggesting the two cell-types link binding affinity and synaptic size similarly.

While the relationship between synaptic size and receptor kinetics was similar between cell types, there were important differences in synaptic organization nonetheless. PMNs had larger glycinergic synapses than SMNs and, associated with that, slower overall receptor turnover. This observation was consistent with previous physiological measurements of inhibitory currents from larval zebrafish MNs that indicate that the lowest input-resistance neurons (very likely PMNs) have larger inhibitory currents [39]. Cell type differences in synaptic organization could have been a consequence of the earlier birth of PMNs and a potentially longer period for the synapses to develop [32]. However, our comparisons of the two cell types in fish of different ages to match neuronal age indicate that this is not the case (Fig. S6). Rather, synapses in both PMNs and SMNs increase in volume with synaptic age, but evidently reach a smaller ultimate size in SMNs (Fig. 7B). A previous study of the dynamics of dendritic filopodia in the two cell types demonstrated lower dynamics in SMNs that were elevated by reducing their activity [24], suggesting differences in synapse formation in the two cell types related to their activity levels. The differences in the size distribution of glycinergic synapses and the associated kinetics likely reflect the functional roles and patterns of activity of the cells within their respective circuits rather than developmental time per se.

Plasticity likely plays an important role in tuning inhibitory synapses within spinal locomotor circuits. Consistent with this idea, glycinergic synapses on zebrafish motoneurons exhibit compensatory homeostatic plasticity due to a mutation resulting in increased extracellular glycine [23,40]. Both activity-dependent [10,14,16,41,42] and homeostatic synaptic scaling [9,10,15–17] have been linked to changes in kinetics of synaptic receptors as well as scaffolding proteins, and when examined, seem to precede changes in synaptic strength [10,14,17]. The ability to measure kinetics in vivo will allow for exploration of their changes during various forms of plasticity.

Our work raises the question of how the distribution of synaptic volumes and kinetics are related to functional aspects of circuits. The distributions of synaptic kinetics observed here may represent specific tuning of contacts to pre-motor networks. In the case of PMNs, the reduction of off kinetics may contribute to potent synapses where robust and precise inhibition is needed. Some of the largest synapses in PMNs were located on the axon initial segment, where critical, potent glycinergic synapses from specialized interneurons in escape circuits connect to PMNs, but not other motoneurons [25–27]. The maintenance of large synapses points to tuning of turnover rates, as without specific control of either off or on kinetics, we might expect a net loss of receptors from large synapses because efflux is proportional to size (see similar argument in [8] ), and because diffusion rates are faster here than binding and unbinding [10]. Similarly specialized large inhibitory synapses were not observed in SMNs, in which kinetics were faster and synaptic volumes smaller. Large synapses at locations other than the initial segment in the PMNs may also represent inputs from distinct sets of pre-synaptic interneurons. Therefore, the distribution of synaptic kinetics and sizes may reflect the differing roles each motoneuron plays within the locomotor network.

Experimental Procedures

More detailed information and procedures are available in Supplemental Experimental Procedures.

Photo-Conversion Experiments

Larvae expressing dendra2-GlyRα1 in isolated MNs were temporarily paralyzed in 0.5 mg/mL curare for 2–3 minutes before being embedded on their sides in 1.6% low melting-point agarose.

Imaging time courses were on the order of 3–4 hours for the majority of experiments. In experiments lasting more than 4 hours, larvae were removed from the agarose in between images and allowed to swim freely.

All imaging was accomplished using a custom-built multi-photon microscope controlled by the ScanImage software package for Matlab (Mathworks). A 960 nm excitation beam (24–40 mW) from a Coherent Chameleon laser was used for excitation. Data were collected using an Apo LWD 25× 1.1 NA water immersion lens (Nikon). Images were collected on 3 PMT channels spectrally filtered into 484–520 nm, 520–562 nm, and 562+ nm. Individual image planes were acquired in half-micron z-steps at a rate of 1.25 Hz at a resolution of 0.0685 μm per pixel.

In order to measure the kinetics of receptor turnover, synaptic puncta were targeted using a 20–30 second 2.5 mW UV laser pulse from a 405 nm Coherent Cube. Puncta typically required 2–3 temporally spaced pulses (30 s apart) to minimize bleaching while generating a sufficient amount of conversion to red fluorescence, and produced noticeable conversion within a 2 μm radius. Several targets were chosen per cell to obtain a range of spot sizes for analysis. In most cases, motoneurons were imaged several times for 2–3 hours. However, some motoneurons were imaged for substantially longer amounts of time, ranging from 6 to 30 hours.

Physiology

The procedure for whole-cell patch recording of muscle fibers is described in Supplemental Experimental Procedures.

Data Processing and Analysis

Fluorescence Recovery after Photo-Conversion

Raw images were converted into an Imaris (Bitplane) time series using custom Matlab (Mathworks) scripts. Both converted and non-converted puncta were identified using Imaris’ spot-detection algorithm, confirmed by visual inspection, and subsequently tracked through time using custom Matlab software. Because manual correction was often required to track synapses between image stacks, we did not track every non-converted puncta. The summed voxel intensity of a 1.5×1.5×4 μm³ cuboid volume centered at each puncta was computed for each spot and the three color channels linearly unmixed after background subtraction using an empirically determined spectral composition (based on spectrally pure non-converted dendra2 and converted dendra2 excited at a 2P emission of 1040 nm, as the green form of dendra2 showed little fluorescence at this excitation) to form a non-converted (green, G) and converted (red, R) fluorescence intensity for each spot through time. To compute relative intensity, G at each time was divided by the mean intensity of all tracked non-converted puncta at the same time. Intensities were scaled such that pre-conversion values were 1 and the first post-conversion 0; furthermore, only data from times greater than 10 minutes post-conversion were fit to an exponential recovery curve f(t) = 1 – exp (−t/ τ) so as to allow diffusion of non-bound converted Dendra2-GlyRα1. Similar results were obtained upon the analysis of the decay of normalized red fluorescence (R/R+G) produced during photo-conversion.

Arbor Position

Synapses were mapped onto the pre-conversion dendritic arbors using the Measurement Points function in Imaris and custom Matlab software. The synaptic distance to soma was calculated as the length in μm taken along dendrites in 3-D space from the soma to that particular synapse.

Synapse Volume

In order to compute synaptic volumes, the pre-conversion image for each cell was de-convolved using an empirically derived point-spread function using AutoQuant X3 Deconvolution software (Media Cybernetics). The de-convolved volume was thresholded at 6σ above the mean fluorescence of dendritic processes, which contain only background levels of dendra2-GlyRα1. Subsequently, the synaptic volume was calculated as the volume of above-threshold pixels within the same 1.5×1.5×4 μm³ region used to compute summed intensity.

Co-localization of Gephyrin and GFP-GlyRα1

The procedure used for immunohistochemistry labeling gephyrin and GFP-GlyRα1 and subsequent co-localization analysis is described in Supplemental Experimental Procedures.

Statistics

All statistics were computed in either Matlab or the R software package. Results in the text are reported with S.E.M when applicable.

Linear Mixed Effects Models

In order to analyze the relationships of multiple parameters to receptor turnover kinetics or synaptic volume in a data set with multiple measurements from the same cell, we used linear mixed effects models. The model coefficients (β_i) describe the change in response variable per unit change of predictor variable holding other factors constant. Linear mixed effects models have a fixed effects component (the independent predictor variables) and a random effects component associated with correlated sources of data (in this case individual motoneurons) that allow modeling of cell-specific slopes and offset as an additional source of structured error. Significant factors in LME models are reported by estimated coefficients (β_i), standard errors, and P-Values. For more details on the use and construction of LME models please refer to Supplementary Experimental Procedures.